Konkani Sentiment Analyzer

Custom Model Training: Fine-tuned transformer-based models on a hand-curated Konkani dataset using PyTorch and Hugging Face.
Labeling Automation: Semi-automated sentiment labeling using Azure Translation + TextBlob to generate initial tags.
Data Preprocessing: Cleaned and structured raw Konkani newspaper text with POS tag removal and CSV formatting.
Evaluation Metrics: Evaluated model using F1-score, accuracy, and confusion matrix across multiple configurations.
Deployment Ready: Model deployed as a web API using FastAPI with testable endpoints and sample inputs.
Language Preservation: Contributes to low-resource language AI and regional NLP by providing a foundational model for Konkani text sentiment.